Jointly analyzing transcriptome, chromatin accessibility and methylation from the same single cells requires principled integration. We develop a high-order kernel spectral clustering method that constructs modality-specific kernels, optimizes a multi-modal Laplacian matrix, and clusters in the shared low-dimensional space. Experiments on real datasets recover known cell types with improved NMI and ARI scores over existing methods, illustrating the power of kernel fusion for multi-omics single-cell analysis.